Re: [PATCH 1/2] am -3: allow nonstandard -p<num> option

[Jeff King <peff@peff.net>]

On Tue, Feb 28, 2012 at 03:24:54PM -0800, Junio C Hamano wrote:

> When falling back to 3-way merge, we run "git apply" to synthesize the
> fake ancestor tree by parsing the incoming patch, and another "git apply"
> to apply the patch to the fake ancestor tree.  Both invocation need to
> be aware of the custom -p<num> setting to parse patches that were prepared
> with non-standard src/dst prefix.

Makes sense. One question:

> diff --git a/git-am.sh b/git-am.sh
> index 1c13b13..d5d168f 100755
> --- a/git-am.sh
> +++ b/git-am.sh
> @@ -127,15 +127,18 @@ fall_back_3way () {
>      mkdir "$dotest/patch-merge-tmp-dir"
>  
>      # First see if the patch records the index info that we can use.
> -    git apply --build-fake-ancestor "$dotest/patch-merge-tmp-index" \
> -	"$dotest/patch" &&
> +    cmd="git apply $git_apply_opt --build-fake-ancestor" &&
> +    cmd="$cmd "'"$dotest/patch-merge-tmp-index" "$dotest/patch"' &&
> +    eval "$cmd" &&

$git_apply_opt can have other stuff in it, too (from my cursory reading,
it looks like --whitespace, --directory, --exclude, -C, --reject,
--ignore-whitespace, and --ignore-space-change).  Those options are now
passed, too.

Naively, I don't think it should be a problem. Many of them will do
nothing (because the patch _should_ apply cleanly to the blobs it
mentions). Some seem like an obvious improvement (e.g., "--directory"
should be just as necessary as "-p", I would think). For something like
"--whitespace=error", I would think we would have errored out already
when we first tried to apply the patch. Or maybe not. I didn't test.

-Peff